Arbitrating and servicing polychronous data requests in Direct Memory Access
View Patent ↗Systems for servicing the data and memory requirements of system devices. A DMA engine that includes a data reservoir is provided that manages and arbitrates the data requests from the system devices. An arbitration unit is provided that only allows eligible devices to make a data request in any given cycle to ensure that all devices will be serviced within a programmable time period. The data reservoir contains the data buffers for each channel of each device. A memory interface ensures that sufficient data for each channel is present in the data reservoir by making requests to a system's memory based on an analysis of each channel. Based on this analysis, a request is either made to the system's main memory, or the channel waits until it is evaluated again in the future. Each channel is thereby guaranteed a response time.
1. For a system that comprises a main memory storing data for one or more devices, each device having one or more associated channels, and a direct memory access (DMA) engine with a data reservoir for transferring data between the one or more devices and the main memory, a computer program product comprising one or more computer readable storage media with computer executable instructions that implement a method of arbitrating access to main memory through the DMA engine, the method comprising the acts of:
providing a device interface to transfer data between the DMA engine and the one or more devices by performing acts of:
selecting at least one of the one or more devices as being eligible to request data from the DMA engine during a particular DMA cycle; and
allowing the at least one eligible device to make a data request to the DMA engine during the particular DMA cycle; and
providing a memory interface to transfer data between the DMA engine and main memory by performing acts of:
defining a circular list having a plurality of entries, each entry corresponding to one of the one or more channels for each device;
evaluating an entry in the circular list to determine whether the channel corresponding to the entry requires service by performing acts of:
determining an entry time that represents how long until the channel will be evaluated again, based at least on how many entries are in the circular list and how much time is needed to service each entry;
determining a latency that represents how long the main memory will take to respond to a request from the DMA engine;
determining a buffer time that represents how long data stored in the data reservoir for the channel will last; and
determining that the channel requires service if the buffer time is less than the entry time and the latency; and
servicing the channel by requesting channel data from the main memory to replenish the data reservoir if the channel requires service.
2. A computer program product as defined in claim 1 , wherein at least one of the plurality of entries in the circular list represents a call to a circular sub list having one or more sub entries, each of which corresponds to one of the one or more channels for each device, the method further comprising an act of evaluating only one of the one or more sub entries in the circular sub list before returning to evaluate the plurality of entries in the circular list.